Methodological quality and bias in randomised controlled trials.

To evaluate the methodological quality of randomised trials in recently published articles and to examine the associations between methodological quality and bias, three related investigations were undertaken. First, to ensure the development of useful measures for the adequacy of randomisation, approaches to allocation were assessed as reported in 206 parallel group trials published in recent volumes of journals of obstetrics and gynaecology. Next, a study was conducted of associations between methodological quality and treatment effects. The material analysed came from 250 trials in 33 meta-analyses on pregnancy and childbirth topics. Finally, the reported approaches to blinding and handling of exclusions were assessed from a random sample of 110 of the 206 previously identified reports. In the 206 published trials, 77% reported either inadequately or unclearly concealed treatment allocation. Additional analyses suggest that non-random manipulation of comparison groups may have occurred. In the next study. compared with trials in which authors reported adequately concealed treatment allocation, trials in which authors reported inadequately or unclearly concealed allocation yielded larger estimates of treatment effects (p<0.001). Odds ratios were distorted by 41% and 33%, respectively. Those associations likely represent bias and are particularly disconcerting in light of the results above from recently published trials. Lack of double-blinding in trials was also associated with larger treatment benefits. However, trials in which authors reported excluding 2 participants after randomisation were not associated with larger treatment effects. That lack of association appeared to be due to incomplete reporting. 3 The analysis of 110 recently published trials also supported the findings that some of the trials not reporting exclusions may actually have had exclusions. In practice, that incomplete reporting could lead to misinterpretations of trial quality. Moreover, only about half the trials that could have double-blinded actually did so. When investigators attempted double-blinding, only 16% provided any written assurances of successfully implementing blinding and only 6% tested its efficacy.